Author + information
- Received November 5, 2001
- Revision received September 20, 2002
- Accepted October 25, 2002
- Published online April 2, 2003.
- Thomas A LaVeist, PhD*,* (, )
- Melanie Arthur, PhD†,
- Athol Morgan, MD, MHS‡,
- Michael Rubinstein, MD∥,
- Joanne Kinder, RN§,
- Linda M Kinney, MPA* and
- Stephen Plantholt, MD§
- ↵*Reprint requests and correspondence:
Dr. Thomas A. LaVeist, Johns Hopkins Bloomberg School of Public Health, 624 N. Broadway, Baltimore, Maryland 21205, USA.
Objectives We sought to identify factors contributing to racial disparity in the receipt of coronary angiography (CA).
Background Numerous studies have demonstrated that African American patients are less likely to receive needed diagnostic and therapeutic coronary procedures than white patients. This report summarizes the methods and findings of a study linking medical records with patient and physician interviews to address racial disparities in the utilization of CA.
Methods This is a retrospective, cross-sectional study conducted in three urban hospitals in Maryland. A total of 9,275 medical records were reviewed, representing all 7,058 cardiac patients admitted in a two-year period. We identified 2,623 patients who, according to American College of Cardiology guidelines, were candidates for receiving CA. A total of 1,669 patients (721 African Americans and 948 whites) and 74% of their physicians were successfully interviewed. Multivariate and hierarchical multivariate logistic regression were used to construct a model of receipt of CA within one year of the hospitalization.
Results The unadjusted odds of white patients receiving CA was three times greater than the odds for African American patients (odds ratio [OR] 3.0, 95% confidence interval [CI] 2.4 to 3.7). Adjusting for patients’ clinical and social characteristics resulted in a 13% reduction in the OR for race. Adjusting for physician and health care system characteristics reduced the OR by 43%, to 1.7 (95% CI 1.3 to 2.4).
Conclusions Racial disparity in the utilization of CA is a function of differences in the health care system “context” in which African American and white patients obtain care, combined with differences in the specific clinical characteristics of patients.
There is a large body of literature documenting racial disparities in the receipt of tertiary-care cardiovascular surgical procedures (1–36). This literature provides overwhelming evidence that clinically appropriate African American patients receive invasive diagnostic and therapeutic procedures commonly prescribed to treat cardiovascular disease less frequently than white patients. Even after adjusting for barriers such as income, type of insurance, access to hospitals with appropriate facilities, and geographic region, white patients are still about twice as likely as African American patients with similar symptom profiles to receive coronary angiography (CA), coronary artery bypass graft surgery, and percutaneous transluminal coronary angioplasty.
Although there is a well-developed body of literature that addresses disparities in the use of these procedures, few previous studies have advanced our understanding much beyond descriptions. This is due primarily to data limitations, as most previous analyses have used medical records or insurance claims data, which contain only limited patient data and typically no information about the physician who managed the patient’s care. A few studies contain data on physicians, but do not link those data with patients’ social or clinical data (33). Additionally, studies that interviewed patients have not linked patient survey responses with clinical data and physician survey data (35,36).
The Cardiac Access Study represents a unique new source of information on this subject. In addition to hospital records used in previous studies, we have interviewed patients and their physicians and linked each source of data. Using this rich source of information, we are able to construct more detailed models of the receipt of CA. The objective of this report is to describe the Cardiac Access Study and summarize the initial findings.
Data source and study population
The Cardiac Access Longitudinal Study is an ongoing study of medical care access, utilization, and quality of life among white and African American cardiac patients from three community hospitals in Baltimore, Maryland. The hospitals are located in close proximity to each other, and there is substantial overlap in their patient base, medical staff, and admitting physicians. The hospitals serve a largely inner-city community; however, the catchment area also extends into suburban areas. An additional advantage of this study setting is that the catchment area served by the hospitals is racially balanced, including urban whites and suburban African Americans. Each hospital has an active cardiology department; however, one of the hospitals did not have a cardiac catheterization laboratory.
We abstracted hospital records to identify patients who were appropriate candidates for receiving CA. We then conducted a telephone survey of the identified CA candidates. Finally, we conducted a mail survey of the physicians who managed the care of the interviewed patients during their hospitalization.
Medical record abstraction
We examined the hospital records of every patient admitted to the hospitals during a two-year period from 1995 to 1997, with a primary diagnosis consistently suggestive of cardiovascular/atherosclerotic disease (see April 2, 2003 JACCissue online at http://www.cardiosource.com for DRGs used to select patients into the study). The set of diagnoses was selected with the goal of including a comprehensive list of all likely diagnoses for which CA may have been appropriate. Although we abstracted a broad range of diagnostic related groups (DRGs), the majority of patients (85.9%) came from a smaller set of DRGs, specifically circulatory disorders, including acute myocardial infarction (AMI) (20.6%), congestive heart failure (31.0%), arrhythmia (9.3%), angina pectoris (8.1%), atherosclerosis (5.4%), and chest pain (11.3%). No other DRG comprised more than 5% of cases.
We identified 9,863 discharges, 9,275 of which were abstracted. Of the remaining charts, 465 (5%) could not be located in the medical records departments, and 123 (1%) were located but were too incomplete to be abstracted. The 9,275 abstracted charts represented 7,058 patients after adjustment for multiple admissions. Trained reviewers abstracted each patient record and classified each patient as class 1, 2, or 3 for receiving CA, according to the criteria established by the American College of Cardiology (ACC) and the American Heart Association (AHA) (37). These guidelines incorporate patients’ symptom severity, noninvasive test results, and other known disease characteristics to assess the appropriateness of referral for CA. According to the ACC/AHA guidelines, class 1 patients (n = 2,282 [32%]) are those for whom there is general agreement that CA is indicated. Class 2 patients (n = 341 [5%]) are those for whom CA is frequently performed but over whom there was a divergence of opinion in the ACC/AHA panel that established the guidelines. For class 3 patients (n = 4,435 [63%]), CA is not indicated. In the case of patients with multiple admissions, the most recent hospitalization for which the patient met class 1 or 2 criteria was considered the focal admission.
Abstractors were not blinded to the patients’ race, as it was impractical to remove all references to race from the more than 9,000 records abstracted. However, a three-member panel of board-certified cardiologists reviewed a random sample of charts for reliability. The panel agreed with the original reviewer in 108 (95%) of 114 cases (kappa = 0.863).
All class 1 and 2 patients (n = 2,623) were followed up by telephone 12 months after discharge. Fifteen percent of the patients were deceased by follow-up, and an additional 8% could not be located. Of the patients who were located alive, 80% (n = 1,669) completed the interview. Reasons for non-completion of the interview included patient refusal (10% of the 2,623 located patients), completion of an interview by proxy due to disability on the part of the patient (5%), unavailability of a proxy for a disabled patient (1%), and partial completion of an interview (<1%). The average length of the interview was 39 min. There was no significant difference by race in the interview length or in the number of refusals to be interviewed.
We conducted a survey of physicians who managed patient care during the hospitalization of the surveyed patients. We selected physicians who were listed on the hospital record as the attending physician or as a consulting cardiologist. We also surveyed physicians who were reported by the patient as the physician who was their “main physician” during the hospitalization. The physician survey (n = 371) was conducted by mail. There were three follow-ups (two by mail and a third by telephone). Twenty-seven physicians (7%) could not be located, either because the names provided by patients were unrecognizable or because physicians had retired or relocated out of the area. A total of 185 physicians returned a completed questionnaire (54% response rate). Responding physicians accounted for 1,239 (74%) of the interviewed patients because most of the non-responding physicians saw only a small number of patients.
For 89% of interviewed patients who were able to identify their “main” physician, we linked that physician if his or her identity could be verified. In the remaining cases, the consulting cardiologist was linked to the patient if a cardiologist had consulted during that patient’s hospitalization. Otherwise, the attending physician from the hospital records was linked to the patient.
The dependent variable is whether or not the patient received CA (1 = received; 0 = not received). The receipt of CA was derived from medical records (if noted). However, we also asked patients if their physicians had discussed the procedure with them and if they had subsequently received it. Patients were asked: “During your visit to [hospital] in [date], did a doctor tell you that you needed heart catheterization? This is when a small tube (also called a catheter) is put into a vein in your leg and passed all the way to your heart. If so, did you receive it?”
Patients with either a notation in the medical record or a positive response to this question were coded as having received a coronary angiogram.
Patient-level independent variables
Patient-level variables are classified into two categories: clinical characteristics and social characteristics. Clinical characteristics include a primary diagnosis, the level of non-cardiac disability, and the patient’s ACC/AHA classification. Social characteristics include race, gender, age, marital status, insurance type, and socioeconomic status.
Patients’ race, primary diagnosis, gender, and age were collected from hospital records and verified by the respondent’s self-report during the telephone interview. The survey protocol called for removing from the study patients who reported their race as other than African American (black) or white. A patient’s race was specified as a binary variable indicating white race. A primary diagnosis was specified as a set of binary variables indicating AMI, congestive heart failure, chest pain or angina, or other. Non-cardiac disability was measured using seven activities of daily living (ADLs). If a patient reported an inability to perform any of these ADLs, we then asked if this was due to his or her heart condition. The ADLs that could not be performed, for reasons other than heart condition, were summed and entered into the models as a continuous variable ranging from 0 to 7. Gender was a binary variable (1 = male; 0 = female), and age was a set of binary variables (<50, 50–64, 65–79, and 80 years and above).
The ACC/AHA classification was ascertained from the hospital record abstraction. It was specified in the analysis as a binary variable (1 = class 1; 0 = class 2). Health insurance was a set of binary variables: private payer, Medicaid, Medicare only, and uninsured. Patients who reported coverage by both Medicare and Medicaid were coded as Medicaid, and patients who had Medicare and private insurance (e.g., Medigap) were coded as private insurance. Marital status was obtained in the patient interview and was coded as currently married or cohabiting versus unmarried, divorced, or widowed.
The patients’ current income, education, and occupational prestige were derived from the patient interview. For analytic purposes, the three variables were collapsed into a single socioeconomic status index, using methods outlined by Nam and Powers (38). First, a patient’s reported occupation (or former occupation if unemployed or retired) was assigned a prestige rating (39). Then, income and education were assigned scaled values. Finally, these three components (income, education, and occupational prestige) were averaged to yield a socioeconomic index score ranging from 0 to 100.
Physician/health system–level independent variables
Variables for the physician/health system–level analysis came from the physician survey and the hospital record abstraction. Physician specialty was a binary variable indicating whether a patient’s main physician during hospitalization was a cardiologist. A physician’s board certification was a binary variable ascertained from the physician survey. One of the study hospitals did not have catheterization facilities; their patients were referred to other hospitals for the procedure. We included a binary variable in the analysis, indicating whether or not a patient was admitted to a hospital with an on-site catheterization laboratory.
Two samples were analyzed: 1) all 1,669 patients who completed an interview; and 2) the subset of 1,239 patients whose physician returned a survey. At each stage, baseline characteristics hypothesized to be related to receiving CA were assessed using the chi-square and Student ttests. To identify predictors of the receipt of CA, we constructed multiple logistic regression models. In analyses that included both patient and physician/health system variables, hierarchical logistic regression models were estimated using the software package HLM5 (40). This technique simultaneously assesses the contribution of patient and physician/health system characteristics to the probability of receiving CA. Hierarchical logistic regression models are preferred over conventional logistic regression because of their ability to account for the clustering of patients among physicians and their greater precision in computing standard errors (40–43).
In Table 1, with respect to several measures, we compared patients who completed the patient interview with non-responding patients to determine whether there was a systematic response bias. Table 1shows that of 2,623 eligible patients, African American patients, males, and older patients were less likely to complete the interview. However, 15% of eligible patients were deceased by the time of the telephone interview (18% and 13% of African American and white patients, respectively; risk ratio [RR] 1.38), and 4% of patients had relocated and could not be found (6% and 2% of African American and white patients, respectively; RR 3.0). Deaths and relocations accounted for 67% of the African American and 58% of the white patients who were not interviewed. We further examined non-response after adjusting for deaths and relocations. In this analysis, we found no significant racial, gender, or ACC/AHA classification differences in interview completion. However, patients over 80 years of age were significantly less likely to complete the interview.
Table 2compares the characteristics of African American patients relative to white patients in the study sample. African American patients differed somewhat from white patients in their primary diagnosis and demographic and socioeconomic characteristics. They were not, however, any less likely to be class 1 candidates for CA. African American patients also differed from white patients in the structural characteristics of their health care, with less access to specialty care, board-certified physicians, and hospitals with cardiac catheterization facilities.
Table 3shows the association between patient-level variables and the receipt of CA. Table 3shows that African American patients had a lower odds of receiving CA than white patients (odds ratio [OR] 2.99, 95% confidence interval [CI] 2.42 to 3.70). Each of the clinical and sociodemographic characteristics were associated with receiving CA in the expected direction. In models 1–3, we estimated a series of multivariate logistic regression models, entering variables in three blocks. This enabled us to examine the effect of each set of variables on the OR for patient race. Model 1 added the patients’ clinical characteristics to the analysis. Adjusting for all three clinical characteristics resulted in an OR of 2.6 (95% CI 2.1 to 3.2) for race. This is a 13% reduction in the effect of race on the receipt of CA, compared with the bivariate OR. Model 2 added the remaining patient-level demographic variables and showed a slight increase in the OR for race. Additional analyses (not shown) indicated that this increase is the result of adjustment for patient age. African American patients in this sample were generally younger than white patients. Model 3 added patients’ socioeconomic status variables, returning the OR to 2.6 for race.
In Table 4, we examined the relationship between each physician/health system–level variable and the receipt of CA. Table 4shows that three of the four variables have a significant bivariate relationship. Only the number of years a physician had been in practice was not significantly associated with procedure use. This variable was not included in subsequent multivariate models.
Models 4 and 5 used hierarchical logistic regression models to add the physician/health system variables to the patient-level variables. In model 4, the addition of physician specialty and board certification reduced the OR for race by 23%, compared with the final patient-level model (model 3 of Table 3). Further adjusting the analysis to include whether or not the patient was seen at a hospital with in-house catheterization facilities (model 5) reduced OR by race to 1.7 (95% CI 1.3 to 2.4), a further 15% reduction of the OR for race. Taking models 4 and 5 together, adjusting for physician and health system variables ultimately reduced the OR for race by 35%, compared with the model with only patient-level predictors (model 3).
We conducted an analysis of data from the Cardiac Access Longitudinal Study, a new data source designed to examine patient and provider characteristics that explain racial disparities in the receipt of CA. Our analysis contained detailed controls for patient characteristics, including socioeconomic status and a variety of provider characteristics. We were able to explain a substantial portion of the racial disparity. In the bivariate model, we obtained an OR of 3.00 (p < 0.001) for patient race, whereas in the final model, the OR was reduced by 43% to 1.7 (p < 0.001).
These findings are summarized in Figure 1, which shows the percentage of racial disparity in the utilization of CA that we were able to explain by patient-level and physician/health care system variables. It appears that the racial disparity in the receipt of CA is, at least partly, a function of differences in the “context” in which African American and white patients obtain care, combined with differences in the specific clinical characteristics of the patients.
Fifty-seven percent of the racial disparities in CA remain unexplained in our analysis. We cannot rule out that there are some race differences in patient preferences or physician attitudes. Analysis of the role of physician race was limited by small numbers of white patients who were seen by African American physicians. Additional limitations of this study are racial differences in those lost to follow-up (due to a higher number of deaths among African Americans), the possibility of racial differences in the urgency of conditions at hospital presentation (which we were unable to account for), the inability to blind medical record abstractors to patient race, and generalizability. Despite these limitations, our findings are consistent with previous findings in the published literature and are robust.
Health care system context
We found a large OR for receiving care at a hospital with an in-house catheterization laboratory. Being required to transfer to another hospital for CA is an important barrier to receiving the procedure. However, the greater likelihood of African American patients to be seen at such hospitals does not account for the racial differences in receiving the procedure. It is also instructive that a large racial disparity in the receipt of CA remained after controlling for patients’ socioeconomic status and insurance coverage.
Physician characteristics are important predictors of the utilization of CA. Auerbach et al. (44)demonstrated that African American patients were significantly less likely to obtain specialty cardiac care. This finding was replicated in these data. Further analysis showed that access to specialty care was an important predictor of the receipt of CA. Also, physician board certification was an important predictor of the receipt of CA, and African Americans were less likely to see a board-certified physician.
These findings are consistent with mounting evidence that racial disparities in the receipt of CA are less a matter of patient preference or other patient characteristics (3,13,18,24,26–38,45)and more a matter of health system characteristics. However, further research on this question is needed. Indeed, although our analysis has accounted for 43% of the racial disparity in CA, this leaves a residual unexplained disparity that remains to be addressed (Appendix).
Drgs used to select patients in study
Descriptions of drgs
115. Permanent cardiac pacemaker implant with AMI, heart failure, or shock
116. Permanent cardiac pacemaker implant without AMI, heart failure, or shock
117. Cardiac pacemaker revision, except for device replacement
118. Cardiac pacemaker device replacement
119. Vein ligation and stripping
120. Other circulatory system or procedures
121. Circulatory disorders with AMI and cardiovascular complications; patient discharged alive
122. Circulatory disorders with AMI without cardiovascular complications; patient discharged alive
123. Circulatory disorders with AMI; patient expired
124. Circulatory disorders, except AMI, with cardiac catheterization and complex diagnosis
125. Circulatory disorders, except AMI, with cardiac catheterization without complex diagnosis
126. Acute and subacute endocarditis
127. Heart failure and shock
129. Cardiac arrest, unexplained
132. Atherosclerosis with complicating condition
133. Atherosclerosis without complicating condition
135. Cardiac congenital and valvular disorders in patients >17 years of age, with complicating condition
136. Cardiac congenital and valvular disorders in patients >17 years of age, without complicating condition
137. Cardiac congenital and valvular disorders in patients 0 to 17 years old
138. Cardiac arrhythmia and conduction disorders with complicating condition
139. Cardiac arrhythmia and conduction disorders without complicating condition
140. Angina pectoris
141. Syncope and collapse with complicating condition
142. Syncope and collapse without complicating condition
143. Chest pain
144. Other circulatory system diagnoses with complicating condition
145. Other circulatory system diagnoses without complicating condition
☆ This research was supported by a grant from the St. Agnes Foundation; grant R01-HL59621 to Dr. LaVeist from the National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland; the Merck Company Foundation; and an unrestricted educational grant from Merck & Co., Inc., Whitehouse Station, New Jersey.
- American College of Cardiology
- activities of daily living
- American Heart Association
- acute myocardial infarction
- coronary angiography
- confidence interval
- diagnostic related group
- odds ratio
- risk ratio
- Received November 5, 2001.
- Revision received September 20, 2002.
- Accepted October 25, 2002.
- American College of Cardiology Foundation
- Funk M.,
- Ostfeld A.M.,
- Chang V.,
- Lee F.A.
- Gillum R.F.,
- Gillum B.S.,
- Francis C.K.
- Goldberg K.D.,
- Hartz A.J.,
- Jacobsen S.,
- Krakauer H.,
- Rimm A.A.
- Mirvis D.M.,
- Burns R.,
- Gaschen L.,
- Cloar F.T.,
- Graney M.
- Okelo S.O.,
- Taylor A.L.,
- Wright J.T.,
- Gordon N.,
- Mohan G.,
- Lesnefsky E.J.
- The American College of Cardiology
- Nam C.B.,
- Powers M.G.
- Nakao K.,
- Treas J.
- Raudenbush S.,
- Bryk A.,
- Cheong Y.F.,
- Congdon R.
- Kreft I.,
- de Leeuw J.
- Bryk A.,
- Raudenbush S.
- Snijders T.,
- Bosker R.
- Auerbach A.D.,
- Hamel M.B.,
- Califf R.M.,
- et al.